Adaptive robot interactions with incremental learning

In order for robots to be adaptable enough to interact effectively they will need to perform some kind of learning that extends their knowledge after deployment. Incremental learning has many advantages relevant to human-robot interaction, including the ability to; learn with little data, train and test simultaneously, update a solution without adversely affecting the old knowledge, and not requiring the system to go offline whilst we perform learning. This paper outlines some incremental learning techniques and how they have been applied to robotic tasks, including how they are applied with our own dasiamedication review robotpsila project.